Multi-Dimensional Temporal Abstraction and Data Mining of Medical Time Series Data: Trends and Challenges

This paper presents emerging trends in the area of temporal abstraction and data mining, as applied to multidimensional data. The clinical context is that of Neonatal Intensive Care, an acute care environment distinguished by multi-dimensional and high-frequency data. Six key trends are identified and classified into the following categories: (1) data; (2) results; (3) integration; and (4) knowledge base. These trends form the basis of next-generation knowledge discovery in data systems, which must address challenges associated with supporting multi-dimensional and real-world clinical data, as well as null hypothesis testing. Architectural drivers for frameworks that support data mining and temporal abstraction include: process-level integration (i.e. workflow order); synthesized knowledge bases for temporal abstraction which combine knowledge derived from both data mining and domain experts; and system-level integration.